Clemson University TigerPrints

Publications Plant and Environmental Sciences

2001 A Partitioned Likelihood Analysis of Swallowtail Butterfly Phylogeny (Lepidoptera: Papilionidae) Michael S. Caterino Clemson University, [email protected]

Robert D. Reed University of Arizona

May M. Kuo University of California - Berkeley

Felix A H Sperling University of California - Berkeley

Follow this and additional works at: https://tigerprints.clemson.edu/ag_pubs Part of the Biology Commons

Recommended Citation Please use publisher's recommended citation.

This Article is brought to you for free and open access by the Plant and Environmental Sciences at TigerPrints. It has been accepted for inclusion in Publications by an authorized administrator of TigerPrints. For more information, please contact [email protected]. Syst. Biol. 50(1):106–127, 2001

A Partitioned Likelihood Analysis of Swallowtail Buttery Phylogeny (Lepidoptera: Papilionidae)

MICHAEL S.CATERINO,1,3 ROBERT D.REED,2 MAY M.KUO,1 AND FELIX A. H. S PERLING1,4 1Division of Insect Biology, 201 Wellman Hall, University of California, Berkeley, CA 94720-3112 , USA 2 Department of Molecular and Cellular Biology, University of Arizona, Life Science South 444, Tucson, AZ 85721, USA

Abstract.—Although it is widely agreed that data from multiple sources are necessary to conŽdently resolve phylogenetic relationships, procedures for accommodating and incorporating heterogeneity in such data remain underdeveloped. Weexplored the useof partitioned, model-based analysesofhetero- geneous molecular data in the context of a phylogenetic study of swallowtail butteries (Lepidoptera: Papilionidae). Despite substantial basic and applied study, phylogenetic relationships among the ma- jor lineages of this prominent group remain contentious. We sequenced 3.3 kb of mitochondrial and Downloaded from nuclear DNA (2.3 kb of cytochrome oxidase I and II and 1.0 kb of elongation factor-1®, respectively) from 22 swallowtails, including representatives of Baroniinae, Parnassiinae, and Papilioninae, and from several moth and buttery outgroups. Using parsimony, we encountered considerable difŽculty in resolving the deepest splits among these taxa. We therefore chose two outgroups with undisputed relationships to each other and to Papilionidae and undertook detailed likelihood analyses of alter- native topologies. Following from previous studies that have demonstrated substantial heterogeneity in the evolutionary dynamics among process partitions of these genes, we estimated evolutionary http://sysbio.oxfordjournals.org/ parameters separately for gene-based and codon-based partitions. These values were then used as the basis for examining the likelihoods of possible resolutions and rootings under several partitioned and unpartitioned likelihood models. Partitioned models gave markedly better Žts to the data than did unpartitioned models and supported different topologies. However, the most likely topology varied from model to model. The most likely ingroup topology under the best-Žtting, six-partition GTR 0 model favors a paraphyletic Parnassiinae. However, when examining the likelihoods of alternativeC rootings of this tree relative to rootings of the classical hypothesis, two rootings of the latter emerge as most likely. Of these two, the most likely rooting is within the Papilioninae, although a rooting be- tween Baronia and the remaining Papilionidae is only nonsigniŽcantly less likely. [Data partitioning; heterogeneity; likelihood; process partitions.] at Clemson University on June 18, 2014

Phylogeny reconstruction is one of the ally necessitates sampling multiple indepen- most dynamic and challenging pursuits in dent sources ofdata (e.g., molecules and mor- modern biology. With recent computational phology, multiple unlinked loci). However, advances, phylogeneticists are increasingly the evolutionary dynamics of independent able to incorporate knowledge of molecular data may vary widely (Bull et al., 1993; Reed evolutionary dynamics in the estimation of and Sperling, 1999), such that a single evo- organismal phylogenies. This becomes par- lutionary model might be inappropriate for ticularly important when examining deeper such heterogeneous data sets. Rather, invok- branches of the tree of life, because with ing several models may be advantageous, sufŽcient time, molecular evolution tends to each one closely matching the dynamics overwrite its own signal, thereby obscuring of one or more of the particular process much phylogenetic information. Maximum partitions of the data (Li`o and Goldman, likelihood methods, which incorporate mod- 1998; Amrine and Springer, 1999; DeBry, els of molecular evolution, can compensate 1999). In this study we examine the perfor- for unobserved substitutions and thus offer a mance of a partitioned likelihood analysis practical solution to this problem. Develop- in reconstructing phylogenetic relationships ing a sound phylogenetic hypothesis gener- among the subfamilies and tribes of papil- ionid butteries. Swallowtail butteries, in the family Pa- 3Present address (and address for correspondence): pilionidae, are among the best known Department of Entomology, The Natural History Mu- insects. Besides serving as the agships of seum, Cromwell Rd., London SW7 5BD, UK; E-mail: histerid@ nhm.ac.uk invertebrate conservation (Collins and 4Present address: Department of Biological Sci- Morris, 1985), swallowtails have been well- ences, CW-405 Biological Sciences Center, University of studied taxonomically and ecologically and Alberta, Edmonton, Alberta T6J 2E9, Canada. have been popular as paradigm systems 106 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 107 for illustrating numerous biological a “living fossil” (Collins and Morris, 1985). phenomena, including mimicry (Clarke The position of Baronia as basal within Pa- and Sheppard, 1963), coevolution (Ehrlich pilionidae seemed assured (Munroe, 1961; and Raven, 1964), and key adaptations Hancock, 1983), but the comprehensive mor- (Berenbaum et al., 1996). A thorough under- phological analysis of buttery phylogeny standing of these evolutionary phenomena by de Jong et al. (1996) has suggested, in- requires a reasonable estimate of phylogeny. stead, that Parnassius might occupy this posi- For example, much of the continued debate tion. Baronia does resemble one of the oldest regarding insect/plant coevolution (e.g., known fossil butteries, Praepapilio colorado Miller, 1987a; Pellmyr et al., 1996; Brower, (Eocene: 48 million years before the present 1997; Farrell and Mitter, 1998) rests on (MaBP); Durdon and Rose, 1978). However, disagreements over phylogenetic details. the resemblance offers no evidence of its phy- Recent studies have made progress in logenetic placement. Even the interpretation

understanding relationships within limited of Praepapilio as a true papilionid has not been Downloaded from groups of Papilionidae (Troidini: Miller, universally accepted (Scott, 1986). Further- 1987b; Weintraub, 1995; Morinaka et al., more, some authorities place the divergence 1999; Battus: Racheli and Oliverio, 1993; of the major swallowtail groups before the Ornithoptera: Parsons, 1996; Papilionini: Gondwanan breakup (i.e., 90 MaBP; Tyler Aubert et al., 1999; Caterino and Sperling, et al., 1994), well before the» time of Praepa- 1999; Reed and Sperling, 1999). However, the pilio. These inconsistencies remain to be http://sysbio.oxfordjournals.org/ higher-level relationships of swallowtails reconciled. The phylogenetic placement of remain equivocal (Rothschild and Jordan, Baronia has important implications for un- 1906; Ford, 1944; Ehrlich, 1958; Munroe, derstanding much of buttery evolution, in 1961; Hancock, 1983; Miller, 1987b; Brown particular, whether its use of a leguminous et al., 1995; Yagi et al., 1999). host represents the plesiotypic buttery con- The Papilionidae contains three subfam- dition (Scott, 1986). ilies: the Baroniinae, Parnassiinae, and Pa- The subfamily Parnassiinae contains 48 pilioninae. The monophyly of the family species in two tribes: the Parnassiini, con-»

is undisputed and is supported by sev- taining the extant genera Archon, Hypermnes- at Clemson University on June 18, 2014 eral synapomorphies (see Kristensen, 1976; tra, and Parnassius (containing 32 of the Hancock, 1983; Miller, 1987b), most convinc- 48 species of Parnassiinae); and the Zeryn- ingly, the larval osmeterium, an eversible, thiini, with Sericinus, Allancastria, Zeryn- forked gland in the thorax that produces thia, Bhutanitis, and LuehdorŽa. H¨auser (1993) and advertises defensive chemicals (Eisner pointed out several weaknesses in the hy- and Meinwald, 1965). The phylogeny of pothesis of parnassiine monophyly, empha- Hancock (1983; our Fig. 1), although not sizing that several uniting features of the universally accepted in all of its details, subfamily actually vary substantially among represents the prevailing hypothesis of sub- the genera, with Hypermnestra, in particu- familial and tribal relationships, and we re- lar, lacking many parnassiine apomorphies. fer to it hereafter as the “classical” hypoth- H¨auser (1993) also noted that the produc- esis. The position of the family within the tion of an elaborate sphragis (a mating Papilionoidea has been controversial. A sis- plug, produced by the male, but observed ter group relationship between Papilionidae on mated females) does not correspond and Pieridae has long been favored (e.g., to the current tribal division, being found Ehrlich, 1958; Scott, 1986). However, placing only in Parnassius, Bhutanitis, and Luehdor- the Papilionidae as the sister lineage to all Ža. H¨auser concluded that even the removal other Papilionoidea is gaining favor (de Jong of the obviously controversial Hypermnestra et al., 1996; Weller et al., 1996). from Parnassiinae would yield a “nonmono- The Baroniinae contains only Baronia bre- phyletic taxon,” a view supported by the vicornis. Populations of this buttery occur morphological studies of de Jong et al. (1996) across southern Mexico in deciduous scrub and by the work of Yagi et al. (1999) on mi- forest where its sole host plant, Acacia cochli- tochondrial NADH dehydrogenase subunit acantha (Fabaceae), occurs (Tyler et al., 1994). 5 (ND5) sequences. On the basis of morphology, Baronia has been The Papilioninae is by far the largest sub- suggested to be the sister lineage to all other family of Papilionidae, with >500 species Papilionidae, and some have referred to it as (Collins and Morris, 1985). Although most 108

FIGURE 1. The “classical” hypothesis of swallowtail higher relationships. This phylogeny is essentially that of Hancock (1983), as represented by the species in this study.

Relationships within Papilionini differ slightly from Hancock’s, following Caterino and Sperling (1999) and Reed and Sperling (1999).

Downloaded from from Downloaded http://sysbio.oxfordjournals.org/ at Clemson University on June 18, 2014 18, June on University Clemson at 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 109 authors agree on its monophyly the rela- 1995), troidines and parnassiines share sev- tionships of the three main tribes (Graphi- eral other seemingly independent charac- ini [ Leptocircini], Troidini, and Papilionini) teristics including, in at least some species, haveD been the subject of considerable contro- asymmetrical tarsal claws, the secretion of versy. (The enigmatic Teinopalpini is gener- a large, visible mating plug (sphragis) by ally placed in Papilioninae as well, but be- the male, and an elongate sclerotized aedea- cause we have not been able to sample this gus (H¨auser, 1993). All of these features have group, their relationships are not discussed been treated variously as symplesiomor- here.) Most authors also have agreed on the phies or convergences, and the issue, as “primitive” nature of the Graphiini, although noted by H¨auser (1993), is as yet unresolved. they have represented this by several differ- In this study we have sampled members ent cladistic hypotheses. Munroe and Ehrlich of most currently recognized papilionid sub- (1960) suggested that the Graphiini might families and tribes (following Miller, 1987b)

be paraphyletic with respect to both the in an effort to resolve these issues. Although Downloaded from Papilionini and the Troidini. Hancock (1983) we were unable to sample a few interesting appeared to propose a sister group relation- genera (i.e., Teinopalpus, Meandrusa, Hyper- ship between Troidini and Papilionini (a rela- mnestra), their absence should not substan- tionship weakly supported by Caterino and tially affect our efforts to examine major phy- Sperling, 1999). However, although Graphi- logenetic events in the family. Using nuclear ini appears monophyletic in Hancock’s phy- (elongation factor-1® [EF-1®]) and mitochon- http://sysbio.oxfordjournals.org/ logeny,his text suggests that it is paraphyletic drial (cytochrome oxidase I and II [COI and with respect to a Troidini Papilionini lin- COII]) protein-coding DNA sequences, we eage (Hancock, 1983:12). TwoC recent molecu- have attempted to determine the higher phy- lar studies have reached conclusions at odds logeny ofthe Papilionidae. The data collected with either of these hypotheses. Yagi et al. also permit some assessment of the relation- (1999) found a sister group relationship be- ship of Papilionidae to other butteries. tween Graphiini and Troidini by using ND5, Given the broad range of divergences in- whereas Morinaka et al. (1999), using the volved in this problem, we recognized from

same gene, found Battus to be more closely the outset that a strict parsimony approach at Clemson University on June 18, 2014 related to Graphiini than to the remaining with the selected genes might prove in- Troidini and considered Graphiini Battus adequate to resolve the deeper nodes. In Papilionini together to constituteC the sisterC a previous study on species-level relation- group to the Troidini. ships within Papilio, Reed and Sperling (1999) The resolution of relationships among examined the relative phylogenetic perfor- the tribes of Papilioninae will have direct mance of these loci. The COI and COII data bearing on the reconstruction of several were found to compromise the resolving intriguing morphological and behavioral power of EF-1® for the deeper nodes of the similarities shared by the Parnassiinae and tree because of homoplasy in the mitochon- Troidini. All Troidini and most genera of drial third codon positions (downweighting Parnassiinae feed exclusively on Aristolochi- of the putatively homoplasious positions im- aceae, storing and using aristolochic acids as proved bootstrap support for these deeper defensive chemicals (von Euw et al., 1968; branches). This assertion was further sup- Rothschild, 1972; Nishida et al., 1993). This ported by estimating the rates of evolution mode of feeding and defense is correlated of gene- and codon-based process partitions with the presence, in the larva, of raised, (sensu Bull et al., 1993) by maximum like- frequently red, tubercles. Ehrlich and Raven lihood; rates among the codon positions of (1964) suggested that Aristolochiaceae feed- the different genes varied as much as 22-fold ing is plesiomorphic within the family (or at (see Table 3 in Reed and Sperling, 1999). least for a common ancestor of Parnassiinae For the purposes of phylogenetic analysis, and Papilioninae). Igarashi (1984) proposed their results suggest that applying a sin- a direct ancestry of an Aristolochia-feeding gle evolutionary model across all the data parnassiine (Sericinus) to the entire Troidini would lead to biased estimates of the ex- in the clearest hypothesis of homology of pected divergence for much of the data. This this habit. Although this conclusion has been problem would cause particular difŽculty in disputed (Miller, 1987a, 1987b; Weintraub, the reconstruction of deep nodes, where the 110 SYSTEMATIC BIOLOGY VOL. 50 accurate estimation of nucleotide divergence pilionoidea. The Hesperiidae (skippers) are is especially troublesome. DeBry (1999) has widely held to be the sister group of the Pa- recently demonstrated that partitioned mod- pilionoidea, or true butteries, and we have els may Žt heterogeneous data better than sequenced representatives of two different unpartitioned models and may, in addition, subfamilies. We Žnally included representa- support alternative topologies. In this study tives of Žve moth families as well as one rep- we therefore have undertaken analyses de- resentative of the enigmatic Hedylidae, long signed to accommodate evolutionary hetero- considered a geometroid moth but now pos- geneity observed among subsets of the data tulated to be a basal buttery (Scoble and by using partitioned likelihood analyses. Aiello, 1990).

Genes MATERIALS AND METHODS We have sequenced the entire mitochon- Ingroup Taxa drial COI and COII genes and 1,000 bp of Downloaded from Our sampled taxa include multiple repre- the nuclear protein-coding EF-1»® gene, for a sentatives of all of the major tribes of Pa- total of 3,328 bp. The deepest papilionid di- pilionidae (Table 1) At the tribal level, we vergences are thought to date to >50 MaBP lack only representatives of the Teinopalpini (Miller, 1987b) with the divergence among (Papilioninae; generally considered to con- buttery families dating to perhaps 80 MaBP http://sysbio.oxfordjournals.org/ tain two genera, Teinopalpus and Meandrusa). or earlier (Scott, 1986). Both of these estimates The subfamily Baroniinae includes only a are based on the few fossil butteries known single species, of which we have examined in concert with the biogeography of extant two individuals. From the Parnassiinae we species. There is little consensus regarding lack representatives of Hypermnestra, Archon, appropriate genes for this range of diver- and Bhutanitis; however, the last of these gences. Two factors have led to our select- is considered closely related to LuehdorŽa ing the mitochondrial genes. First, because (Hancock, 1983), which is examined here. a substantial database of lepidopteran se- From the Papilionini we have included three quences already exists for these genes, these genera of Graphiini, Žve genera of Troi- data are a valuable asset to studies of the at Clemson University on June 18, 2014 dini, and six species from widely separated evolution of these genes as well as to the species groups of Papilio (Papilionini.) All of prospect of a global lepidopteran phylogeny. the sequences of Papilionini and one each And second, though COI and COII are con- of the Troidini and Graphiini were used in sidered to be relatively quickly evolving at the previous studies of Caterino and Sperling the nucleotide level, and therefore may con- (1999) and Reed and Sperling (1999). tain substantial homoplasy, we Žnd com- pelling Hillis’s (1996) suggestion that suf- Outgroup Taxa Žcient sampling density can overcome this problem. The nuclear protein-coding gene Because the root of the Papilionidae is un- EF-1® has been evaluated by Cho et al. (1995) clear, we sequenced a wide variety of Lepi- and Mitchell et al. (1997), who demonstrated dopteran outgroups. The general consensus informativeness of synonymous nucleotide has been that the Pieridae is the closest rel- substitutions up to divergences of 60 million ative of the Papilionidae (e.g., Scott, 1985), years (main branches of Noctuoidea) and and we thus include one member each of two postulated deeper resolution with increas- pierid subfamilies. However, recent morpho- ingly dense taxon sampling. logical (de Jong et al., 1996) and combined data (Weller et al., 1996) studies have sug- gested that Papilionidae may be the sister Molecular Techniques group of the remaining Papilionoidea. Ac- Total genomic DNA was extracted as in cording to this view, any other Papilionoidea Sperling and Harrison (1994) or using a Qia- might serve as appropriate outgroups; there- gen QIAamp tissue kit. Polymerase chain re- fore, we also included sequences from one action (PCR) ampliŽcations were performed member of each of Nymphalidae, Satyri- with either an Ericomp TwinBlock EasyCy- dae, Riodinidae, and Lycaenidae. We also in- cler or an MJ Research PTC-200 DNA Engine cluded several taxa from outside of the Pa- and using a hot start: Taq was added at the 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 111

TABLE 1. Species sampled, with localities and GenBank accession numbers.

GenBank accession no. Species Locality COI-COII EF-1® Noctuidae aFeltia jaculifera (pheromone type A) CAN: AB U60990 AF173390 Geometridae bLambdina Žscellaria CAN: NF AF064521 AF173391 Sphingidae Proserpinus clarkiae USA: CA AF170855 AF173394 Saturniidae Hemileuca electra USA: CA AF170856 AF173395 Hedylidae Macrosoma sp. Costa Rica AF170854 AF173393 Pyralidae Ostrinia nubilalis CAN: ON AF170853 AF173392 Downloaded from Hesperiidae Pyrginae Erynnis tristis USA: CA AF170858 AF173397 Pyrgus communis USA: CA AF170857 AF173396 Hesperiinae Hylephila phyleus USA: CA AF170859 AF173398 http://sysbio.oxfordjournals.org/ Satyridae Coenonympha tullia USA: CA AF170860 AF173399 Nymphalidae Boloria epithore USA: CA AF170862 AF173402 Riodinidae Apodemia mormo USA: CA AF170863 AF173403 Lycaenidae Euphilotes bernardino USA: CA AF170864 AF173404 Pieridae cColias eurytheme CAN: ON AF044024 AF173400 Pieris napi USA: CA AF170861 AF173401

Papilionidae at Clemson University on June 18, 2014 Baroniinae Baronia brevicornis (two specimens) Mexico AF170865 AF173405 AF170866 AF173406 Parnassiinae Parnassiini Parnassius clodius (simo group) USA: WA AF170871 AF173411 Parnassius phoebus (apollo group) CAN: AB AF170872 AF173412 Zerynthiini Allancastria cerisy Greece AF170869 AF173409 LuehdorŽa japonica Japan: Kanazawa AF170867 AF173407 Zerynthia rumina Spain: Malaga AF170870 AF173410 Sericinus montela Japan: near Tokyo AF170868 AF173408 Papilioninae Graphiini Graphium (Graphium) agamemnon SE Asia AF170874 AF173414 Iphiclides podalirius France AF170873 AF173413 c Eurytides (Protesilaus) marcellus USA: FL AF044022 AF044815 Troidini Troides (Troides) helena Malaysia AF170878 AF173418 Battus philenor USA: VA AF170875 AF173415 Atrophaneura alcinous Japan: Okura AF170876 AF173416 Parides photinus Costa Rica AF170877 AF173417 c Pachliopta neptunus Malaysia AF044023 AF044829 Papilionini c Papilio (Pterourus) glaucus USA: MD AF044013 AF044826 c Papilio (Pterourus) troilus USA: FL AF044017 AF044820 c Papilio (Papilio) machaon France: Coudoux AF044006 AF044819 c Papilio (Heraclides) cresphontes USA: WI AF044004 AF044832 c Papilio (Princeps) xuthus Japan: Tokyo AF043999 AF044838 c Papilio (Princeps) demoleus Malaysia AF044000 AF055825

aSperling et al., 1996. bSperling et al., 1999. cCaterino and Sperling (1999), Reed and Sperling (1999). 112 SYSTEMATIC BIOLOGY VOL. 50 end of an initial denaturation at 94±C; this fects of weighting based on a priori (co- was followed by 35 cycles of 1 min at 94±C, don positions and transition/transversion 1 min at 45±C, 1.5 min at 72±C and a sub- weighting) and a posteriori (reweighting by sequent 5-min Žnal extension at 72±C. PCR rescaled consistency indices [RCI]) criteria. products were cleaned by using a Qiagen Weighting ultimately made only small differ- PCR PuriŽcation Kit and then were cycle- ences for papilionid resolution (see below). sequenced with Perkin-Elmer/ABI Dye Ter- Support for branches under parsimony was minator Cycle Sequencing Kit with Ampli- assessed by bootstrap analyses (1,000 repli- taq FS on an MJ Research PTC-200 according cates starting with simple stepwise addi- to Perkin-Elmer’s suggested thermal proŽle. tion trees, TBR branch swapping). Decay Sequenced products were Žltered through indices were also calculated for selected anal- Sephadex-packed columns and dried. Se- yses. Minimum evolution trees were con- quencing was performed with an ABI 377 structed by using Jukes–Cantor (JC; Jukes

automated sequencer. All fragments were and Cantor, 1969), Kimura two-parameter Downloaded from sequenced in both directions. Sequences (K2P; Kimura, 1980), Hasegawa–Kishino- were aligned manually to the sequences Yano (HKY85; Hasegawa et al., 1985), and of Drosophila yakuba (COI/COII; Clary and LogDet (Steel, 1994) distances. Wolstenholme, 1985) or Heliothodes diminu- Given the best-supported ingroup topolo- tivus (EF-1®; Cho et al., 1995). Most primers gies derived from the preceding analyses, used are published in Caterino and Sperling we examined likelihoods of alternative hy- http://sysbio.oxfordjournals.org/ (1999) and Reed and Sperling (1999). Addi- potheses under several models. Likelihoods tional primers used are given in Appendix 1. were calculated under the JC, K2P, HKY85, HKY85 0, and General Time Reversible (GTR; LanaveC et al., 1984) 0 models over Phylogenetic Analysis the entire unpartitioned dataC set. The nec- DNA sequences were aligned by eye, with essary model parameters were estimated use of translated amino acid sequences in over each topology for each model. We also the few instances of length variation. All calculated likelihoods under three of these

phylogenetic analyses were performed with models—the JC, HKY85 0, and GTR 0— at Clemson University on June 18, 2014 C C beta test versions of PAUP¤ (4b2–4b4a; Swof- over a six-partition data set. The designated ford, 1999). At the outset, we partitioned the partitions were (1) COI/COII Žrst codon po- nucleotide and amino acid sequence data sitions, (2) COI/COII second codon posi- into mitochondrial and nuclear subsets and tions, (3) COI/COII third codon positions, examined them for incongruence, using the (4) EF-1® Žrst codon positions, (5) EF-1® sec- Incongruence Length Difference test (ILD; ond codon positions, and (6) EF-1® third Farris et al., 1994) in PAUP¤. It has been codon positions. The low number of variable suggested that the ILD test is an overly sites for the Žrst and second codon positions conservative estimator of combinability of EF-1® may pose problems for parameter (Cunningham, 1997). Therefore, despite estimation (high estimate variance). How- some indications of incongruence (see ever, given the low rates of change at these Results, below), parsimony analyses were positions, they are expected to provide im- performed on the entire nucleotide data portant information for basal relationships, set as well as on the separate genes. These and we have chosen to maintain them as preliminary results indicated good resolving separate partitions. The tRNA-leucine and power for relationships within Papilionidae intergene spacers of the mitochondrial se- but limited informativeness with respect quences were excluded from likelihood cal- to outgroups. We treated the ingroup sep- culations (because existing likelihood mod- arately for most analyses and considered els do not accommodate gaps well). To obtain the problem of rooting the ingroup tree in log-likelihoods for partitioned models, log- subsequent analyses. likelihoods were calculated for each partition Ingroup analyses proceeded from heuris- independently and then summed. Model pa- tic parsimony searches (100 random taxon rameters for partitioned models (® for a addition replicates, TBR branch-swapping, four-category approximation to a gamma gaps scored as missing data) with use of distribution, transition/transversion ratios, equally weighted separate and combined nu- and substitution rate matrices) were inde- cleotide data sets. We also examined the ef- pendently estimated and optimized on Žxed 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 113 topologies for each model and partition. In ceeded from the assumption that all buttery all cases, the observed nucleotide frequencies outgroups were approximately equally in- for each individual partition were used. Ac- formative (given a model that can accurately tual parameter values may be obtained from account for unobserved substitutions, any the senior author. The Kishino–Hasegawa outgroup should be approximately as useful test (Kishino and Hasegawa, 1989), as im- as any other outgroup). We selected two plemented in PAUP¤, was used to test for unrooted ingroup topologies for examining signiŽcant differences in likelihood among rooted likelihoods: the classical hypothesis, topologies for each unpartitioned model, al- and that favored by the most parameter- though the conŽdence intervals for this test rich model examined here (six-partition may be too narrow for comparing more than GTR 0). We then grafted two outgroup two topologies (Shimodaira and Hasegawa, taxa—oneC from the Pieridae (Pieris), fre- 1999). For partitioned models it was only pos- quently considered the sister group to

sible to directly test for differences among Papilionidae, and one from the Hesperiidae Downloaded from topologies within a particular partition. (Pyrgus), which is clearly outside of the An important advantage of likelihood Papilionoidea—to several possible branches methods is that they can be used to ex- and calculated the likelihoods under the amine the assumptions underlying the evo- unpartitioned and partitioned HKY85 0 lutionary models used (Goldman, 1993; and GTR 0 models. All model parametersC Huelsenbeck and Crandall, 1997; Huelsen- were againC estimated and optimized for http://sysbio.oxfordjournals.org/ beck and Rannala, 1997). We used likelihood each model and partition. Likelihood dif- ratio tests (LRTs) to test for statistically sig- ferences among topologies and among niŽcant differences in model Žt for models partitions were tested with Kishino–Hase- with increasing complexity. Given that likeli- gawa tests. hoods were calculated over Žxed topologies, the models used may be treated as nested RESULTS hypotheses and the distribution of the LRT statistic (twice the difference between the two Data Properties 2

likelihoods) is expected to approximate a  The Žnal nucleotide data set contained at Clemson University on June 18, 2014 distribution (but see Goldman, 1993; Whelan 3,328 positions (2,333 mitochondrial, 995 nu- and Goldman, 1999). The appropriate degree clear; gaps are observed at 68 posi- of freedom for the test is then the difference in tions), translating to 1,069 amino acids (740 the number of free parameters between the mitochondrial, 329 nuclear). Basic variabil- models being compared (Felsenstein, 1981; ity statistics for all sequences are presented Huelsenbeck and Rannala, 1997). in Table 2. These data present a remark- Under parsimony, alternative outgroups able range of divergences among genes and yielded widely differing ingroup rootings, codon positions. The low extreme is repre- with no taxa better supported as a papilionid sented by EF-1® second positions, which had sister group than any other. Therefore, for the a maximum pairwise divergence of <3%, purposes of rooting the ingroup tree, we pro- whereas nearly all third positions exhibited

TABLE 2. Nucleotide variability over genes and codon position partitions, assessed by parsimony reconstruction on one of two most-parsimonious combined data ingroup-only topologies.

All sites Codon pos. 1 Codon pos. 2 Codon pos. 3 Amino acids

mt EF mt EF mt EF mt EF mt EF No. characters 2333 995 738 332 738 331 738 332 738 331 No. invariant 1513 701 559 307 679 326 178 68 581 314 No. variable 820 294 179 25 59 5 560 264 159 15 No. informative 632 242 136 17 39 3 446 222 106 8 Autapomorphies 188 52 43 8 20 2 114 42 53 7 CI 0.407 0.455 0.431 0.448 0.647 0.500 0.384 0.454 0.634 0.444 RI 0.392 0.537 0.479 0.529 0.700 0.286 0.349 0.539 0.647 0.444 Ti/Tv 0.893 2.716 2.641 4.500 0.902 1.25 0.681 2.868 - 114 SYSTEMATIC BIOLOGY VOL. 50 divergences >20%. First and second codon stitutions or compositional biases, or both) positions of all protein-coding genes fall and parsimony weighting schemes (simul- largely within reliable ranges, with mito- taneously downweighting third positions chondrial Žrst positions showing the greatest and transitions by one-half, third position divergence, at 10%. Divergences at the Žrst transitions thereby being weighted by one- two codon positions» are approximately twice fourth) resulted in a monophyletic Papil- as high in the mitochondrial data as in EF-1®, ionidae (results not shown), though rela- indicating greater rates of protein evolution tionships among the outgroup butteries as well as nucleotide evolution. The range of and moths still exhibited improbable rela- third position divergences is greater in EF-1®, tionships (e.g., [[Papilionidae[Hesperiidae ranging from 8% to nearly 50%, whereas remaining Papilionoidea]]). These analysesC few mitochondrial» comparisons exceed 30%. support the idea that the distant comparisons This is almost certainly a result of the highly are hindered by severe homoplasy. Thus,

skewed AT bias of insect mitochondria (% taking papilionid monophyly as supported, Downloaded from A/C/G/T for all positions: COI and COII: based on corrected analyses, we pruned the 32/14/12/42; EF-1® 27/25/25/23). D outgroups and undertook analyses of papil- ILD testing betweenD mitochondrial and ionid taxa alone. nuclear partitions yielded signiŽcant incon- Parsimony searches over the combined gruence between partitions based on nu- equally-weighted nucleotide data for in- cleotides (P 0:01) but not, however, be- group taxa yielded two equally parsimo- http://sysbio.oxfordjournals.org/ tween aminoD acid partitions (P 0:189). One nious trees (3,613 steps; CI 0.4204; RI interpretation of this result isD that homo- 0.4419) (Figs. 3a, 3b). AnalysesD of the sepa-D plasy in the nucleotide data may be ob- rate data sets resulted in two trees for the scuring the phylogenetic signal. However, mitochondrial data (Fig. 3c, 3d) (2,715 steps; only 13 of the EF1-® amino acids are infor- CI 0.3568; RI 0.3925) and Žve trees for the mative under parsimony (considering out- EF-1D® data (Figs.D 3e–3i) (878 steps; CI group ingroup), and congruence between 0.4715; RI 0.5789), all of which were dis-D aminoC acid partitions may result in part tinct. A strictD consensus of these nine trees

from low resolution in the EF-1® partition. is presented in Figure 4. The groups sup- at Clemson University on June 18, 2014 We also conducted ILD tests with only in- ported in all of these (numbered as in Fig. 4) group taxa (Papilionidae), resulting in non- are (1) Baroniinae, (2) Parnassiini (two Par- signiŽcant differences for nucleotide (P nassius species), (3) Zerynthiini without Lue- 0:300) and amino acid partitions (P 0:120).D hdorŽa (Sericinus Allancastria Zerynthia), We suggest that in this case homoplasyD (4) Papilioninae,C (5) Graphiini,C (6) Troidini among the more distant comparisons is mim- without Battus, and (7) Papilionini. Our main icking the effects of incongruence as as- concerns here are relationships among these sessed with ILD. We accept the result of the major clades, and relationships within them more restricted (ingroup only) test and com- will not be addressed further. bine the mitochondrial and nuclear parti- Parnassiine monophyly is consistent with tions. one of two combined-data trees (Fig. 3b) although bootstrap support is <70% (even after RCI reweighting). Mitochondrial nu- Parsimony Analyses cleotides alone suggest that the Zerynthiini Parsimony searches over the full data set (minus LuehdorŽa) may be the sister group with all nucleotides equally weighted re- of the remaining Parnassiinae and Papil- sulted in two equally parsimonious trees ioninae (Figs. 3c, 3d), whereas EF-1® nu- (7,180 steps; CI 0.280; RI 0.352). The strict cleotides suggest that Zerynthiini is more consensus (Fig.D 2) of these revealsD that deeper closely related to Papilioninae than to Par- nodes are poorly resolved; most notably, nassiini (Figs. 3d–3h), although, again, boot- neither Papilionoidea nor Papilionidae was strap support is weak (<50%). As with found to be monophyletic. Baronia appears the analysis of the complete ingroup plus more closely related to two nonswallow- outgroup data, weighting schemes (down- tail butteries and to Hemileuca, a saturniid weighting faster-evolving positions and nu- moth, than to other swallowtails. Distance cleotides) and distance analyses (all models transformations (including K2P, HKY85, and examined [JC, K2P, HKY85, GTR, LogDet]) LogDet, which compensate for multiple sub- designed to compensate for multiple hits 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 115 Downloaded from http://sysbio.oxfordjournals.org/ at Clemson University on June 18, 2014

FIGURE 2. Unrooted strict consensus of two equally parsimonious trees based on all unweighted nucleotides over all ingroup outgroup taxa only (7,180 steps; CI 0.2797; RI 0.3520). Values above the branches indicate bootstrap supportC (where these exceed 50%) and those belowD branchesD indicate Bremer support. nearly all support a monophyletic Parnassi- vant branches have >85% bootstrap sup- inae (results not shown). The placement of port for combined nucleotides. However, LuehdorŽa is equivocal. Based on combined two of Žve EF-1® trees are inconsistent analysis (Figs. 3a, 3b) and mitochondrial with this resolution, placing the Papilion- data alone (Figs. 3c, 3d), the genus is re- ini as sister to Graphiini Troidini in solved as being more closely related to Par- one (Fig. 3h), and with BattusC split from nassius than to Zerynthiini, although without the remaining Troidini in another (Fig. 3i). strong bootstrap support. EF-1®, however, Although the combined data results seem supports the placement of LuehdorŽa with sufŽcient to reject these alternatives, we reex- Zerynthiini (Figs. 3e–3i). This resolution is amined these relationships using likelihood only weakly supported by bootstrapping analyses. (56%), but that increases to 73% when transi- tions are downweighted by one-half, and to 81% when EF-1® nucleotides are reweighted Likelihood Analyses by their CI values. Within Papilionini, the The parsimony-based analyses offer a va- most frequent result is the classical reso- riety of possible resolutions of the major lution of the tribes, with a monophyletic swallowtail lineages, with little basis for Graphiini sister to monophyletic Troidini choosing among them. Likelihood analysis Papilionini (Figs. 3a–3g), and all rele- offers a means of distinguishing among this C 116 SYSTEMATIC BIOLOGY VOL. 50 Downloaded from http://sysbio.oxfordjournals.org/ at Clemson University on June 18, 2014

FIGURE 3. Most-parsimonious trees derived from equally weighted separate and combined analyses: (a, b) All nucleotides included; (c, d) mitochondrial data alone; (e–i) EF-1® data alone. array. Log-likelihoods were calculated over with LuehdorŽa at the base of the Parnassi- all parsimony trees (Figs. 3a–3i) plus two ini rather than with the Zerynthiini. Results additional trees not found among them: the are presented in Table 3, with models ar- classical hypothesis (Fig. 1) and a tree consis- ranged roughly in order of increasing com- tent with a monophyletic Parnassiinae but plexity from left to right. The goodness-of-Žt 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 117

topology EF-1® 2 clearly best Žts these mod- els, and this therefore is the tree we carried forward as the “favored” ingroup topology for rooting purposes. The difŽculty of assessing signiŽcance un- der some models must render any con- clusions based on our likelihood analyses tentative; nonetheless, a couple of ques- tions deserve deeper examination. Firstly,the aberrant resolutions of the tribes of Papilioni- nae found in some most-parsimonious EF-1® topologies (Figs. 3h, 3i) are found here to be least likely under nearly all models, and the

paraphyly of Troidini can be rejected with Downloaded from FIGURE 4. Strict consensus of the nine trees in Fig- ure 3, showing groups found in all equally weighted sep- statistical conŽdence. An additional impor- arate and combined analyses. Numbers above branches tant issue regards the placement of Luehdor- indicate bootstrap support (where >50%) for the mito- Ža within Parnassiinae. Although the parsi- chondrial/nuclear/combined data. Numbers below the mony results are equivocal, with only the branches designate clades referred to in the text. EF-1® trees favoring the classical placement with Zerynthiini, trees containing this resolu- http://sysbio.oxfordjournals.org/ tion are favored under the best-Žtting HKY85 of these models improves substantially with 0 and GTR 0 models, both unparti- increasing parameter-richness; LRTs sup- tionedC and partitioned,C and this is the hy- port all comparisons as highly signiŽcant pothesis we favor. The monophyly of the (results not shown). Particularly large im- Parnassiinae as a whole is more difŽcult to provements in model Žt are seen both with establish. The only trees to contain this group the incorporation of gamma-distributed rate (classical and “combined data 2” [Figs. 1 and heterogeneity and with data partitioning. 3b, respectively]) rank Žrst in unpartitioned

The favored topology varies widely among analyses and in the partitioned JC analysis. at Clemson University on June 18, 2014 models. The simplest unpartitioned models Yet, under the most realistic partitioned mod- favor a topology (Fig. 3b) that differs from els, paraphyly of Parnassiinae appears more the classical hypothesis with regard to both likely, and the question cannot be considered the position of LuehdorŽa (with the Parnassi- resolved. ini) and the resolution of the three graphi- Two topologies were used for likelihood ine taxa. When gamma-distributed rate analyses of root placement: the classical hy- heterogeneity is incorporated into the unpar- pothesis, and that favored by our best-Žtting titioned HKY85 and GTR models, the clas- likelihood model (EF-1® 2). The selected out- sical hypothesis (Fig. 1) is favored in both groups, Pyrgus communis (Hesperiidae) and cases. However, when the data are parti- Pieris napi (Pieridae), were grafted onto the tioned and the model parameters are esti- trees on the branches numbered in Figure 5. mated separately for each partition, this fa- The rootings examined include the classi- vored topology shifts, Žrst back to the same cal Baronia-basal tree (rootings 1 and 8), the combined data hypothesis as was supported Parnassius-basal rooting supported by the by simple unpartitioned models (Fig. 3b), morphological data of de Jong et al. (1996; and then to one in which the Parnassi- rootings 5 and 10), and a range of others to inae appear paraphyletic with respect to provide a context for evaluating the likeli- the Papilioninae (Fig. 3f). Although the log- hood scores. Some, such as those within Pa- likelihoods results are not signiŽcantly dif- pilionini, were expected to be quite unlikely ferent from the classical hypothesis for any at the outset. partition, the “likelihood advantage” (see The calculated likelihoods for the rooted DeBry, 1999) for the best-scoring topology in- topologies are shown in Table 4 (with scores creases slightly with improved model Žt (this for individual partitions in Appendix 2). is despite a decrease in the total range of the As with the ingroup-only calculations, im- estimates). Combined with the nearly iden- provements in model Žt, as assessed with tical rankings of topologies under the parti- LRTs, are all highly signiŽcant (results not tioned HKY85 0 and GTR 0 models, the shown). Interestingly, two rootings of the C C 118

TABLE 3. Log likelihoods of alternative ingroup topologies under various models. The classical tree is that shown in Figure 1. The Lueh. with Parn. tree is the classical tree with LuehdorŽa moved to the base of the Parnassiini. The remaining topologies are shown in Figure 3 and represent all of the most-parsimonious topologies based on equally weighted, separate and combined analyses. The best likelihood scores under each model are shown in boldface type. The number of model parameters for each model is calculated as [data partitions (branch lengths rate parameter ti/tv ratio nucleotide frequencies relative substitution rate parameters rate heterogeneity parameter)]. The likelihoods for individual partitions aCre given in AppenCdix 2. C C C

Unpartitioned Partitioned

Unrooted topology JC K2P HKY HKY 0 GTR 0 JC HKY 0 GTR 0 C C C C Classical tree 23697.227 23483.111 22971.759 20499.228 19939.540 21377.254 18182.537 17948.142 Lueh. with Parn. ¡23699.752 ¡23484.180 ¡22969.076 ¡20500.050 ¡19941.197 ¡21373.563 ¡18188.387 ¡17951.960 Combined data 1 ¡23692.902 ¡23471.435 ¡22957.391 ¡20505.934 ¡19951.141 ¡21374.504 ¡18198.853b ¡17958.293 Combined data 2 ¡23685.511 ¡23466.516 ¡22953.883 ¡20502.020 ¡19947.497 ¡21367.962 ¡18194.649 ¡17957.609 ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 1 23692.902 23471.435 22957.391 20505.934 19951.141 21374.504 18198.853b 17958.293 Mitochondrial 2 ¡23703.700 ¡23480.973 ¡22965.791 ¡20509.133 ¡19957.023 ¡21378.782 ¡18201.374b ¡17962.788b EF-1® 1 ¡23722.898 ¡23511.356 ¡22994.403 ¡20510.722 ¡19946.083 ¡21407.856c ¡18193.996 ¡17954.377 EF-1® 2 ¡23707.488 ¡23493.396 ¡22978.475 ¡20504.673 ¡19940.100 ¡21384.098c ¡18177.096 ¡17941.079 EF-1® 3 ¡23717.193 ¡23503.570 ¡22987.113 ¡20505.314 ¡19941.070 ¡21404.683c ¡18187.896 ¡17948.779 EF-1® 4 ¡23831.073a ¡23617.916a ¡23095.653a ¡20563.750a ¡19995.600a ¡21501.398c ¡18225.053c ¡18014.822c EF-1® 5 ¡23752.553 ¡23540.541 ¡23019.986 ¡20522.631 ¡19957.592 ¡21431.992c ¡18197.657 ¡17969.078 Range of lnL estimates ¡ 145.562 ¡ 151.400 ¡ 141.770 ¡ 64.522 ¡ 56.060 ¡ 133.436 ¡ 47.957 ¡ 73.743 Free parameters 42 43 46 47 53 252 282 318

aSigniŽcantly less likely (at ® 0:05) as determined by Kishino–Hasegawa test. D bOne or more nuclear partitions Žt this topology signiŽcantly worse (at ® 0:05) as determined by Kishino–Hasegawa test (see Appendix 2). D

cOne or more mitochondrial partitions Žt this topology signiŽcantly worse (at ® 0:05) as determined by Kishino-Hasegawa test (see Appendix 2).

D

Downloaded from from Downloaded http://sysbio.oxfordjournals.org/ at Clemson University on June 18, 2014 18, June on University Clemson at 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 119 Downloaded from http://sysbio.oxfordjournals.org/

FIGURE 5. Selected ingroup topologies. The topology on the left is the classical hypothesis. The one on the right is the hypothesis favored by likelihood analyses of ingroup taxa. Candidate rooting points used for calculating at Clemson University on June 18, 2014 rooted likelihoods are numbered as in Table 4. classical hypothesis scored better than any surprising topologies—with scant available rootings of the topology favored by ingroup- criteria for differentiating them. Although only likelihood analyses, underscoring the parsimony encountered problems early on, persistent uncertainty regarding ingroup re- presumably attributable to homoplasy, we lationships. Of the two best-scoring rooted anticipated that likelihood analysis would be topologies, one of the rootings within Papil- able to satisfactorily compensate for it to re- ioninae was unexpectedly found to be most solve the deeper relationships accurately.The likely—under the better Žtting of the two demonstrated heterogeneity in evolutionary unpartitioned models and under both par- dynamics of process partitions of these data titioned models. However, the Baronia-basal offered an opportunity toexamine the perfor- rooting of the classical hypothesis ranks Žrst mance of partitioned models, for which evo- under the unpartitioned HKY85 0 model, lutionary parameters could be separately op- and is a close second by all others.C Neither timized. Given this heterogeneity,the models of these best-scoring topologies is statisti- developed in this paper are both intuitively cally distinguishable from most other topolo- appealing and seemingly accurate. Previous gies, for combined data or for any individual studies have noted that simple models fre- partitions. quently identify the same most-likely topol- ogy as complex ones do (e.g., Cunningham, 1997; DeBry, 1999), the implication being that CONCLUSIONS choice of model, surprisingly, is not as criti- Our analyses have spanned a wide range cal as it would seem. However, our results of tree reconstruction and evaluation strate- indicate that this idea requires further eval- gies. In the end, we are left with an array of uation. Although some consistency was ap- possible resolutions—some suggesting very parent across a range of more or less simple 120 SYSTEMATIC BIOLOGY VOL. 50

TABLE 4. Log likelihoods of alternative rootings of the two selected ingroup topologies (Fig. 5). Ingroup trees were rooted with the outgroups Pyrgus (Hesperiidae) and Pieris (Pieridae). Tree numbers refer to the rooting points in Figure 5. Parsimony scores and log-likelihoods were calculated with outgroups attached, though they are not shown in these schematics. The scores for the best trees are shown in boldface type and those that differ at ® 0:05 are marked with asterisks (for partitioned likelihoods, the asterisks indicate signiŽcance for one or more partitions).D Likelihoods for individual partitions are given in Appendix 3.

Unpartitioned Partitioned

Tree no. Rooted topology MP score HKY85 0 GTR 0 HKY85 0 GTR 0 C C C C Parnassiinae 4141 22521.982 21882.310 20018.460 19762.630 7 Baroniinae ¡ ¡ ¡ ¡ Graphiini Troidini Papilionini Baroniinae 4134 22521.360 21884.347 20020.255 19763.633 1 Parnassiinae ¡ ¡ ¡ ¡ Downloaded from Graphiini Troidini Papilionini Baroniinae 4143 22529.120 21884.157 20028.634 19767.901 Parnassiini ¡ ¡ ¡ ¡ 13 Zerynthiini Graphiini http://sysbio.oxfordjournals.org/ Troidini Papilionini Parnassiinae 4137 22523.847 21885.585 20024.323 19768.668 2 Baroniinae ¡ ¡ ¡ ¡ Graphiini Troidini Papilionini Parnassiinae 4143 22525.187 21886.598 20040.472 19769.883 3 Baroniinae ¡ ¡ ¡ ¡ Graphiini Troidini Papilionini at Clemson University on June 18, 2014 Graphiini 4140 22530.564 21891.242 20029.770 19774.098 6 Parnassiinae ¡ ¡ ¡ ¡ Baroniinae Troidini Papilionini Baroniinae 4145 22536.031 21892.718 20036.998 19774.680 Parnassiini ¡ ¡ ¡ ¡ 8 Zerynthiini Graphiini Troidini Papilionini Baroniinae 4143 22531.601 21888.889 20035.160 19775.292 Parnassiini ¡ ¡ ¡ ¡ 11 Zerynthiini Graphiini Troidini Papilionini

Baroniinae 4148¤ 22536.826 21892.935 20036.916 19776.165 Parnassiini ¡ ¡ ¡ ¡ 9 Zerynthiini Graphiini Troidini Papilionini Parnassiini 4150 22537.972 21893.609 20040.043 19777.839 Baroniinae ¡ ¡ ¡ ¡ 10 Zerynthiini Graphiini Troidini Papilionini Parnassiini 4144 22537.077 21896.544 20056.224 19782.922 Zerynthiini ¡ ¡ ¡ ¡ 5 Baroniinae Graphiini Troidini Papilionini 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 121

TABLE 4. Continued.

Unpartitioned Partitioned

Tree no. Rooted topology MP score HKY85 0 GTR 0 HKY85 0 GTR 0 C C C C Zerynthiini 4148 22540.145 ¤ 21899.139 20042.945 19784.296 Parnassiini ¡ ¡ ¡ ¡ 4 Baroniinae Graphiini Troidini Papilionini

Graphiini 4142 22537.420 21892.946 20039.141 19805.977 ¤ Baroniinae ¡ ¡ ¡ ¡ 12 Parnassiini Zerynthiini Troidini Papilionini Downloaded from models with regard to favored topology, in- discriminate among topologies. However, creasing model complexity caused the pre- in this study, this value might also be re- ferred topology to shift, especially in the case lated to the overall range of likelihood of the rooted analysis, to a strikingly different estimates. phylogeny. Both the changes in preferred topology http://sysbio.oxfordjournals.org/ The major shortcoming of the approach and the possibility of increased discrimina- presented here is the inability to make any tory power point to yet more complex mod- statistical statements regarding the parti- els as a possible salvation. However, this con- tioned analyses. In a few cases, individ- clusion would not be entirely warranted. Par- ual partitions show considerable conict titioning the data decreases the number of with particular topologies. However, the variable characters in each partition, which differences required for signiŽcance under leads to higher variances of parameter and Kishino–Hasegawa tests for unpartitioned likelihood estimates and thence to poten- models (which would be expected to have tially spurious results (Swofford et al., 1996). less variance than the partitioned estimates) The six-partition model presented here may at Clemson University on June 18, 2014 suggest that few, if any partitioned compar- suffer from this difŽculty for some partitions; isons would be signiŽcantly different. This for example, the Žrst and second codon posi- appears to indicate either that the Kishino– tions of EF-1® offer 25 and 5 variable sites, Hasegawa test is insufŽciently sensitive for respectively. In fact, the partitioned analy- detecting real likelihood differences or that ses are not able to extract any phylogenetic the likelihoods of these trees do not ac- information whatsoever from these second tually differ. In fact, we believe that both codon positions (see Appendices 2, 3), and of these factors apply to our results. The the information presented by the EF-1® Žrst close agreement in the rankings of topolo- positions may be similarly suspect. (How- gies across a wide selection of models ar- ever, a four-partition analysis with all EF-1® gues against their likelihoods differing only data combined, carried out at the sugges- through random variation in estimates. Para- tion of one reviewer, resulted in the same metric bootstrapping may be necessary toad- rankings of trees.) Possibly additional par- equately establish the variance of these es- titioning of highly variable partitions (e.g., timates. Nonetheless, many likelihoods for mitochondrial third positions) according to these topologies probably would not differ functional regions, amino acid properties, signiŽcantly by any conceivable test. In the codon biases, and so forth, would extract case of the two most-likely rooted topolo- additional information from these data. Fur- gies, (Table 4), which differ by at most two ther explorations to determine optimal mod- log-likelihood units, the topologies suggest els are needed. radically different evolutionary scenarios, With respect to swallowtail relationships, and this problem merits serious considera- our Žndings concur broadly with accepted tion. The apparent increase in likelihood ad- ideas. However, despite sampling from all vantage, at least for ingroup-only analyses, major taxa of the swallowtail family (and suggests that additional model improve- outgroups), and extensive sequencing from ments would yield additional power to loci that seem appropriate, we obtained 122 SYSTEMATIC BIOLOGY VOL. 50 relatively few results that can be highlighted ably not be affected by the addition of as incontrovertible. Nonetheless, with two the parnassiines Archon (which is also an possible exceptions, we believe that our anal- Aristolochia-feeder) and Hypermnestra (for yses best support a tree congruent with tradi- which its Zygophyllaceae-feeding would be tional classiŽcation. Most important, our in- reconstructed as autapomorphic, wherever group analyses suggest that the Parnassiinae the taxon belongs on the cladogram). Ad- is not monophyletic. A monophyletic Parnas- ditional Papilioninae have the greatest po- siinae is found in only one of nine parsimony tential to provide a new perspective on trees and is not supported by the partitioned Aristolochiaceae-feeding. In particular, if likelihood analysis. Instead, our ingroup basal Papilionini and basal Graphiini are analyses favor a sister group relationship be- found to share similar host-plant families tween Zerynthiini (including LuehdorŽa) and (for the taxa here there is no overlap), the Papilioninae, and morphology would not Aristolochiaceae-feeding in the Troidini will

contradict such a relationship. This resolu- have to be viewed as an autapomorphic de- Downloaded from tion was also found by Yagi et al. (1999), parture from some ancestral Papilioninae using ND5. Secondly, our analysis cannot habit. On the other hand, depending on its conŽdently establish the root of the swallow- phylogenetic placement, Aristolochiaceae- tail tree. Although in this case morphology feeding in the asius group of Protesilaus would conict, a rooting within the Papilion- (Graphiini) could potentially reinforce the inae is as likely as the classical Baronia-basal single origin hypothesis. In any event, too http://sysbio.oxfordjournals.org/ rooting based on our analyses (including few phylogenetic data are available to draw “corrected” parsimony analyses of the full any serious conclusions regarding host-plant outgroup ingroup dataset.) With regard to evolution. previous workers’C hypotheses, we Žnd no In briey summarizing the behavior of support for Munroe and Ehrlich’s (1960) these markers over the range of divergences suggestion of a closer relationship between examined in this study, the most notewor- Graphiini and Papilionini than between Troi- thy point is that, contrary to the widely- dini and Papilionini. A sister group relation- held view that the COI/COII loci are mainly

ship between Troidini and Graphiini, as sug- useful for species-level studies, they are in at Clemson University on June 18, 2014 gested by Yagi et al. (1999), was found in fact much more widely applicable. As has some initial parsimony trees but is strongly been observed previously, the third codon rejected by likelihood analysis. The hypoth- positions of EF-1® do offer information at esis of troidine polyphyly suggested by deeper levels and over a greater range than Morinaka et al. (1999) is not supported by do those of COI/COII (Reed and Sperling, any ofour analyses. Before any of these issues 1999). However, due to the differences in can be considered settled, however, substan- amino acid variability, the Žrst and second tial phylogenetic work remains to be done. codon positions of the mitochondrial data Several problematic genera need to be ex- offer far more informative sites than do amined, most notably the parnassiines Ar- those of a comparable amount of EF-1® se- chon and Hypermnestra and the papilionines quence at the phylogenetic levels examined Teinopalpus and Meandrusa. Evaluation of re- here. Indeed, EF-1® amino acid sequences lationships at this level might also beneŽt have been found useful at far deeper in- from the examination of a nuclear ribosomal terclass levels in Arthropoda (Regier and locus, such as 18S. Schultz, 1997). Our initial hope was that by A thorough exploration of the origin and partitioning these data and estimating phy- evolution of Aristolochiaceae-feeding and logeny by using maximum likelihood, our its associated morphologies and behaviors analysis might extend the utility of the ob- is outside the scope of this study. How- served variation. That we have been suc- ever, a single origin of this feeding mode cessful, however, is not clear. The likelihood does map most-parsimoniously onto either analysis has conclusively settled few of the the classical hypothesis (as represented by ambiguities presented by the parsimony the taxa included here) or the Parnassiinae- analysis. Certainly our arsenal of loci would paraphyletic tree favored by our likelihood beneŽt from the development of addi- analysis (coding each taxon for its known tional single-copy nuclear genes for which host plant family or families, following the amino acid sequences evolved at a Hancock, 1983). This result would prob- higher rate than that of EF-1a. Particularly 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 123

promising candidates are dopa decarboxy- BERENBAUM, M. R., C. FAVRET, AND M. A. SCHULER. lase (Fang et al., 1997; Friedlander et al., 1996. On deŽning “key innovations” in an adaptive radiation: Cytochrome P450s and Papilionidae. Am. 1998) and phosphoenolpyruvate carboxyk- Nat. 148:S139–S155. inase (Friedlander et al., 1996; see also BROWER, A. V. Z. 1997. The evolution of ecologically Friedlander et al., 1992.) important characters in Heliconius butteries (Lepi- Phylogeny reconstruction remains a difŽ- doptera: Nymphalidae): A cladistic review. Zool. J. cult task. Whereas by most accounts denser Linn. Soc. 119:457–472. BROWN, K. S., C. F. KLITZKE, C. BERLINGERI, AND P. taxon sampling should lead to greater EUZEBIO´ RUBBO DOS SANTOS. 1995. Neotropical swal- phylogenetic accuracy (Hillis, 1996, 1998; lowtails: Chemistry of food plant relationships, pop- Graybeal, 1998), perhaps an unanticipated ulation ecology and biosystematics. Pages 405–446 consequence is that as taxa are added and in Swallowtail butteries: Their ecology and evolu- tionary biology (J. M. Scriber, Y. Tsubaki, and R. C. branch lengths decrease, conŽdence in par- Lederhouse, eds.). ScientiŽc Publishers, Gainesville, ticular branches of these complex trees Florida. BULL, J. J., J. P.HUELSENBECK, C. W. CUNNINGHAM, D. L.

will be more difŽcult to assess by conven- Downloaded from tional means. Our results suggest that more- SWOFFORD, AND P.J. WADDELL. 1993. Partitioning and complex evolutionary models may be better combining data in phylogenetic analysis. Syst. Biol. 42:384–397. able to discern differences between similar CATERINO, M. S., AND F. A. H. SPERLING. 1999. Papilio topologies. Phylogenetics therefore stands to phylogeny based on mitochondrial cytochrome oxi- beneŽt from the continued development of dase I and II genes. Mol. Phylogenet. Evol. 11:122– evolutionary models that can account for the 137. http://sysbio.oxfordjournals.org/ CHO, S., A. MITCHELL,J.C.REGIER,C.MITTER, vagaries of heterogeneous data. This goal re- R. W. POOLE, T. P.FRIEDLANDER, AND S.ZHAO. 1995. quires both detailed examinations of the evo- A highly conserved gene for low level phylogenetics: lutionary dynamics of process partitions of Elongation factor-1 alpha recovers morphology-based popular phylogenetic markers and the elabo- tree for heliothine moths. Mol. Biol. Evol. 12:650–656. ration of methods for applying simultaneous CLARKE, C. A., AND P. M. SHEPPARD. 1963. Interactions between major genes and polygenes in the determina- partitioned analyses in software packages for tion of the mimetic patterns of Papilio dardanus. Evo- phylogenetic analysis. Perhaps the most im- lution 17:404–413. portant remaining question is the degree to CLARY, D. O., AND D. R. WOLSTENHOLME. 1985. The mitochondrial DNA molecule of Drosophila yakuba:

which data should be partitioned to opti- at Clemson University on June 18, 2014 Nucleotide sequence, gene organization and genetic mally extract information. No recommenda- code. J. Mol. Evol. 22:252–271. tion beyond Žnding an undeŽnable balance COLLINS, N. M., AND M. G. MORRIS. 1985. Threatened between complexity and statistical practical- swallowtail butteries of the world. IUCN, Gland and ity can be offered at this point. Future work Cambridge. would proŽtably focus on developing crite- CUNNINGHAM, C. W. 1997. Is congruence between data partitions a reliable predictor of phylogenetic accu- ria for identifying this optimal balance for a racy? Empirically testing an iterative procedure for given phylogenetic problem. choosing among phylogenetic methods. Syst. Biol. 46:464–478. DEBRY, R. W. 1999. Maximum likelihood analysis of ACKNOWLEDGMENTS gene–based and structure–based process partitions, In addition to collectors acknowledged in our pre- using mammalian mitochondrial genomes. Syst. Biol. vious papers, we thank R. Kelson, J. Powell, and D. 48:286–299. Rubinoff for providing specimens. We also thank A. DE JONG, R., R. I. VANE-WRIGHT, AND P. R. ACKERY. Engberg and N. Vane for laboratory assistance. This 1996. The higher classiŽcation of butteries (Lepi- manuscript was much improved by the comments of doptera): Problems and prospects. Entomol. Scand. A. Cognato, R. DeBry, A. Mitchell, R. Olmstead, H. True- 27:65–101. man, and one anonymous reviewer. This study was sup- DURDON, C. J., AND H. ROSE. 1978. Butteries from the ported in part by California Agricultural Experiment middle Eocene: The earliest occurrence of fossil Papil- Station and NSF-PEET grants to F.A.H.S. ionidae. Pearce-Sellards Ser. Tex. Mem. Mus. 29:1–25. EHRLICH, P. R. 1958. The comparative morphology, phy- logeny and higher classiŽcation of the butteries. Univ. Kansas Sci. Bull. 39:305–370. REFERENCES EHRLICH, P. R., AND P. H. RAVEN. 1964. Butteries and AMRINE, H. M., AND M.S.SPRINGER. 1999. Maximum- plants: A study in coevolution. Evolution 18:586–608. likelihood analysis of the Tethythere hypothesis based EISNER, T., AND Y. C. MEINWALD. 1965. Defensive secre- on a multigene data set and a comparison of different tion of a caterpillar (Papilio). Science 150:1733–1735. models of sequence evolution. J. Mamm. Evol. 6:161– FANG, Q. Q., S. CHO, J. C. REGIER,C.MITTER, 176. M.MATTHEWS, R. W. POOLE, T. P.FRIEDLANDER, AND AUBERT, J., L. LEGAL, H. DESCIMON, AND F. MICHEL. S.ZHAO. 1997. A new nuclear gene for insect phylo- 1999. Molecular phylogeny of swallowtail butteries genetics: DOPA decarboxylase is informative of re- of the tribe Papilionini. Mol. Phylogenet. Evol. 12:156– lationships within Heliothinae (Lepidoptera: Noctu- 167. idae). Syst. Biol. 46:269–283. 124 SYSTEMATIC BIOLOGY VOL. 50

FARRELL, B. D., AND C. MITTER. 1998. The tim- LANAVE, C., G. PREPARATA,C.SACCONE, AND G. SERIO. ing of insect-plant diversiŽcation: Might Tetraopes 1984. A new method for calculating evolutionary sub- (Coleoptera: Cerambycidae) and Asclepias (Asclepi- stitution rates. J. Mol. Evol. 20:86–93. adaceae) have co-evolved? Biol. J. Linn. Soc. 63:553– LIO` , P., AND N.GOLDMAN. 1998. Models of molecu- 577. lar evolution and phylogeny. Genome Res. 8:1233– FARRIS, J. S., M. KALLERSJ¨ O¨ , A. G. KLUGE, AND C.BULT. 1244. 1994. Testing signiŽcance of incongruence. Cladistics MILLER, J. S. 1987a. Host–plant relationships in the Pa- 10:315–319. pilionidae (Lepidoptera): Parallel cladogenesis or col- FELSENS TEIN, J. 1981. Evolutionary trees from DNA se- onization? Cladistics 3:105–120. quences: A maximum likelihood approach. J. Mol. MILLER, J. S. 1987b. Phylogenetic studies in the Papilion- Evol. 17:368–376. inae (Lepidoptera: Papilionidae). Bull. Am. Mus. Nat. FORD, E. B. 1944. Studies in the chemistry of pigments Hist. 186:365–512. in the Lepidoptera, with reference to their bearing on MITCHELL, A., S. CHO, J. C. REGIER,C.MITTER, R. W. systematics. 4. The classiŽcation of the Papilionidae. POOLE, AND M.MATTHEWS. 1997. Phylogenetic utility Trans. R. Entomol. Soc. London 94:201–223. of Elongation factor-1 alpha in Noctuoidea (Insecta: FRIEDLANDER, T. P., R. HORST-KURT, J. C. REGIER, C. Lepidoptera): The limits of synonymous substitution. MITTER, R. S. PEIGLER, AND Q. Q. FANG. 1998. Two Mol. Biol. Evol. 14:381–390. Downloaded from nuclear genes yield concordant relationships within MORINAKA, S., T. MAEYAMA, K. MAEKAWA, D. Attacini (Lepidoptera: Saturniidae). Mol. Phylogenet. ERNIWATI, S. N. PRIJONO, I. K. GINARSA, T. Evol. 9:131–140. NAKAZAWA, AND T. HIDAKA. 1999. Molecular phy- FRIEDLANDER, T. P., J. C. REGIER, AND C.MITTER. 1992. logeny of birdwing butteries based on the represen- Nuclear gene sequences for higher level phylogenetic tatives in most genera of the tribe Troidini. Entomol. analysis: 14 promising candidates. Syst. Biol. 41:483– Sci. 2:347–358.

490. MUNROE, E. 1961. The classiŽcation of the Papilionidae. http://sysbio.oxfordjournals.org/ FRIEDLANDER, T. P., J. C. REGIER, C. MITTER, AND D. L. Can. Entomol. Suppl. 17:1–51. WAGNER. 1996. A nuclear gene for higher level phylo- MUNROE, E., AND P.R. EHRLICH. 1960. Harmonization of genetics: Phosphoenolpyruvate carboxykinase tracks concepts of higher classiŽcation of the Papilionidae. J. Mesozoic-age divergences within Lepidoptera. Mol. Lepid. Soc. 14:169–175. Biol. Evol. 13:594–604. NISHIDA, R., J. D. WEINTRAUB, P.FEENY, AND H.FUKAMI. GOLDMAN, N. 1993. Statistical tests of models of DNA 1993. Aristolochic acids from Thottea spp. (Aristolochi- substitution. J. Mol. Evol. 36:182–198. aceae) and the osmeterial secretions of Thottea-feeding GRAYBEAL, A. 1998. Is it better to add taxa or characters troidine swallowtail larvae (Papilionidae.) J. Chem. to a difŽcultphylogenetic problem? Syst. Biol. 47:9–17. Ecol. 19:1587–1594. HANCOCK, D. L. 1983. ClassiŽcation of the Papilionidae: PARSONS, M. J. 1996. A phylogenetic reap- A phylogenetic approach. Smithersia. 2:1–48. praisal of the birdwing genus Ornithoptera HASEGAWA, M., H. KISHINO, AND T. YANO. 1985. Dating (Lepidoptera: Papilionidae: Troidini) and a new at Clemson University on June 18, 2014 the human-ape split by a molecular clock of mitochon- theory of its evolution in relation to Gondwanan drial DNA. J. Mol. Evol. 22:160–174. vicariance biogeography. J. Nat. Hist. 30:1707–1736. HAUSER¨ , C. L. 1993. Critical comments on the phylo- PELLMYR, O., J. N. THOMPS ON, J. M. BROWN, AND R. G. genetic relationships within the family Papilionidae. HARRISON. 1996. Evolution of pollination and mutu- Nota Lepid. 16:34–43. alism in the yucca moth lineage. Am. Nat. 148:827– HILLIS, D. M. 1996. Inferring complex phylogenies. 847. Nature 383:130–131. RACHELI, T., AND M. OLIVERIO. 1993. Biogeographical HILLIS, D. M. 1998. Taxonomic sampling, phylogenetic patterns of the neotropical genus Battus Scopoli (Lep- accuracy, and investigator bias. Syst. Biol. 47:3–8. idoptera: Papilionidae.) Trop. Zool. 6:55–65. HUELSENBECK, J. P., AND K. A. CRANDALL. 1997. Phy- REED, R. D., AND F. A. H. SPERLING . 1999. Interaction logeny estimation and hypothesis testing using max- of process partitions in phylogenetic analysis: An imum likelihood. Annu. Rev. Ecol. Syst. 28:437–466. example from the swallowtail buttery genus Papilio. HUELSENBECK, J. P., AND B.RANNALA. 1997. Phyloge- Mol. Biol. Evol. 16:286–297. netic methods come of age: Testing hypotheses in an REGIER, J. C., AND J. W. SHULTZ. 1997. Molecular phy- evolutionary context. Science 276:227–232. logeny of the major arthropod groups indicates poly- IGARASHI, S. 1984. The classiŽcation of the Papilionidae phyly of crustaceans and a new hypothesis for the mainly based on the morphology of their immature origin of hexapods. Mol. Biol. Evol. 14:902–913. stages. Tyo To Ga 34:41–96. ROTHS CHILD, M. 1972. Secondary plant substances and JUKES, T. H., AND C. R. CANTOR. 1969. Evolution of pro- warning colouration in insects. Pages 59–83 in In- tein molecules. Pages 21–132 in Mammalian protein sect/plant relationships (H. F. van Emden, ed.). R. En- metabolism (H. N. Munro, ed.). Academic Press, New tomol. Soc. London Symp. 6. York. ROTHS CHILD, W., AND K. JORDAN. 1906. A revision of KIMURA, M. 1980. A simple method for estimating evo- the American Papilios. Novit Zool. 13:412–752. lutionary rate of base substitution through compar- SCOBLE, M. J., AND A. AEILLO. 1990. Moth-like butter- ative studies of nucleotide sequences. J. Mol. Evol. ies (Hedylidae: Lepidoptera): A summary,with com- 16:111–120. ments on the egg. J. Nat. Hist. 24:159–164. KIS HINO, H., AND M. HASEGAWA. 1989. Evaluation of SCOTT, J. A. 1986. The butteries of North America: A the maximum likelihood estimate of the evolution- natural history and Želd guide. Stanford University ary tree topologies from DNA sequence data, and the Press: Stanford, California. branching order in Hominoidea. J. Mol. Evol. 29:170– SHIMODAIRA, H., AND M.HASEGAWA. 1999. Multi- 179. ple comparisons of log–likelihoods with applications KRISTENSEN, N. P. 1976. Remarks on the family-level to phylogenetic inference. Mol. Biol. Evol. 16:1114– phylogeny of butteries. Z. Zool. Syst. Evol. 14:25–33. 1116. 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 125

SIMON, C., F. FRATI, A. BECKENBACH, B. CRESPI, H. LIU, TYLER, H., K. S. BROWN, AND K. WILSON. 1994. Swal- AND P. FLOOK. 1994. Evolution, weighting, and phy- lowtail butteries of the Americas: A study in bio- logenetic utility of mitochondrial gene sequences and logical dynamics, ecological diversity, biosystematics a compilation of conserved polymerase chain reaction and conservation. ScientiŽc Publishers, Gainesville, primers. Ann. Entomol. Soc. Am. 87:651–701. Florida. SPERLING , F. A. H., R. BYERS, AND D. HICKEY. 1996. Mi- VON EUW, J., T. REICHSTEIN, AND M.ROTHSCHILD. 1968. tochondrial DNA sequence variation among phero- Aristolochic acid in the swallowtail buttery Pach- motypes of the dingy cutworm, Feltia jaculifera (Gn.) lioptera aristolochiae. Isr. J. Chem. 6:659–670. (Lepidoptera: Noctuidae). Can. J. Zool. 74:2109– WEINTRAUB, J. D. 1995. Host plant association patterns 2117. and phylogeny in the tribe Troidini (Lepidoptera: Pa- SPERLING , F. A. H., AND R. G. HARRISON. 1994. Mito- pilionidae). Pages 307–316 in Swallowtail butteries: chondrial DNA variation within and between species Their ecology and evolutionary biology.(Scriber, J. M., of the Papilio machaon group of swallowtail butteries. Tsubaki, Y., and R. C. Lederhouse, eds.). ScientiŽc Pub- Evolution 48:408–422. lishers, Gainesville, Florida. SPERLING , F. A. H., A. G. RASKE, AND I. S. OTVOS. 1999. WELLER, S. J., D. P. PASHLEY, AND J. A. MARTIN. 1996. Mitochondrial DNA sequence variation among popu- Reassessment of buttery family relationships using lations and host races of Lambdina Žscellaria (Gn.) (Lep- independent genes and morphology. Ann. Entomol. Downloaded from idoptera: Geometridae). Insect Mol. Biol. 8:1–10. Soc. Am. 89:184–192. STEEL, M. 1994. Recovering a tree from the Markov WHELAN, S., AND N.GOLDMAN. 1999. Distributions of leaf colourations it generates under a Markov model. statistics used for the comparison of models of se- Appl. Math. Lett. 7:19–23. quence evolution in phylogenetics. Mol. Biol. Evol. SWOFFORD, D. L. 1999. PAUP¤: Phylogenetic Analysis 16:1292–1299. Using Parsimony,ver.4.0b2a. SinauerAssociates, Sun- YAGI, T., G. SASAKI, AND H.TAKEBE. 1999. Phylogeny

derland, Massachusetts. of Japanese papilionid butteries inferred from nu- http://sysbio.oxfordjournals.org/ SWOFFORD, D. L., G. J. OLS EN, P. J. WADDELL, AND D. M. cleotide sequences of the mitochondrial ND5 gene. J. HILLIS. 1996. Phylogenetic Inference. Pages 407–514 Mol Evol. 48:42–48. in Molecular systematics (D. M. Hillis, C. Moritz, and B. K. Mable, eds.). Sinauer Associates, Sunderland, Received 29 October 1999; accepted 13 December 1999 Massachusetts. Associate Editor: R. Olmstead

APPENDIX 1.

In addition to primers listed in Caterino and Sperling (1999), Reed and Sperling (1999), and Cho et al. (1995), we at Clemson University on June 18, 2014 used or designed the following primers for this study. Most are minor variants of existing primers. Mitochondrial location numbers refer to Drosophila yakuba (Clary and Wolstenholme, 1985). Nuclear location numbers refer to Heliothodes diminutivus (Cho et al., 1995). See Simon et al. (1994) for additional mitochondrial primers at these and other sites. Gene F/R Location Sequence COI R 1751 GGA GCT CCA GAT ATA GCT TTC CC R 1840 TGG GGG GTA TAC TGT TCA (T/A) CC R 2329 ACA GTA AAT ATA TGA TGT GCT CA R 2329 ACT GTG AAT ATG TGA TGG GCT CA R 2329 ACA GTA AAT ATA TGA TGA GCC CA F 2495 CCT CTA TAC TTT GAA GAT TAG G F 2495 CAT CAA TT(C/T) TAT GAA GAT TAG G F 2495 CCT CAA TTT TAT GAA GAT TAG G R 3014 TCA TTG CAT TTA TCT GCC ACA TTA COII F 3038 CTA ATA TGG CAG ATT ATA TCT AAT GGA EF-1® F 453 AAC TGA GCC ACC TTA CAG TGA GAG R 551 GGA GAC AAC ATG CTG GAC TCC A R 572 CTC CTT ACG CTC AAC ATT CC R 1048 AAC CGT TTG AGA TTT GAC CAG GG 126 SYSTEMATIC BIOLOGY VOL. 50

APPENDIX 2. Log likelihoods of unrooted ingroup-only topologies for individual data partitions under three separate models. The best likelihoods under each partition and model are shown in boldface type. Asterisks indicate a likelihood signiŽcantly worse than the best for that partition (as determined by Kishino–Hasegawa test; ® 0.05). D COI/II EF-1®

Topology Pos1 Pos2 Pos3 Pos1 Pos2 Pos3 Sum

Jukes–Cantor Standard tree 3979.349 1747.648 10119.523 855.248 528.567 4146.912 21377.254 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Lueh. with Parn. 3983.291 1746.031 10107.738 855.401 528.567 4152.535 21373.563 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Combined data 1 3984.145 1755.120 10093.330 852.967 528.567 4160.373 21374.504 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Combined data 2 3975.927 1748.974 10096.152 858.531 528.567 4159.811 21367.960 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 1 3984.145 1755.120 10093.333 852.967 528.567 4160.373 21374.504 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 2 3983.551 1755.120 10096.576 846.540 528.567 4168.427¤ 21378.782 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Downloaded from EF-1® 1 4003.487 1758.311 10128.958 ¤ 850.714 528.567 4137.820 21407.856 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 2 3984.633 1752.328 10130.972 ¤ 849.996 528.567 4137.602 21384.098 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 3 4001.549 1758.311 10129.517 ¤ 850.079 528.567 4136.659 21404.683 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 4 4038.044 ¤ 1777.139¤ 10171.629 ¤ 840.429 528.567 4145.589 21501.398 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 5 4014.354 ¤ 1762.293 10146.479 ¤ 842.938 528.567 4137.361 21431.992 ¡ ¡ ¡ ¡ ¡ ¡ ¡

HKY85 0 http://sysbio.oxfordjournals.org/ C Standard tree 3396.144 1632.758 8057.197 775.884 514.509 3806.043 18182.537 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Lueh. with Parn. 3397.046 1631.942 8055.900 775.884 514.509 3813.105 18188.387 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Combined data 1 3397.408 1636.982 8054.574 774.388 514.509 3820.991 ¤ 18198.853 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Combined data 2 3395.443 1632.800 8054.867 777.851 514.509 3819.179 18194.649 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 1 3397.408 1636.982 8054.574 774.388 514.509 3820.991 ¤ 18198.853 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 2 3397.079 1636.982 8054.181 770.915 514.509 3827.707 ¤ 18201.374 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 1 3404.945 1640.792 8059.500 771.581 514.509 3802.668 18193.996 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 2 3398.060 1636.012 8058.535 769.807 514.509 3800.173 18177.100 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 3 3404.040 1640.792 8058.574 769.807 514.509 3800.173 18187.896 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 4 3414.287 ¤ 1651.695¤ 8066.857 767.046 514.509 3810.659 18225.053

¡ ¡ ¡ ¡ ¡ ¡ ¡ at Clemson University on June 18, 2014 EF-1® 5 3408.670 1642.468 8061.169 768.287 514.509 3802.553 18197.657 ¡ ¡ ¡ ¡ ¡ ¡ ¡ GTR 0 C Standard tree 3309.198 1615.683 8034.699 735.677 505.767 3747.118 17948.142 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Lueh. with Parn. 3310.100 1615.033 8033.377 735.705 505.767 3751.977 17951.960 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Combined data 1 3309.150 1619.865 8031.845 733.616 505.767 3758.049 17958.293 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Combined data 2 3308.092 1615.913 8032.476 737.683 505.767 3757.677 17957.609 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 1 3309.150 1619.865 8031.845 733.616 505.767 3758.049 17958.293 ¡ ¡ ¡ ¡ ¡ ¡ ¡ Mitochondrial 2 3309.595 1619.865 8031.762 731.543 505.767 3764.255 ¤ 17962.788 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 1 3314.284 1622.761 8036.635 730.416 505.767 3744.514 17954.377 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 2 3309.775 1618.871 8035.885 728.511 505.767 3742.269 17941.080 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 3 3313.578 1622.761 8035.893 728.511 505.767 3742.269 17948.779 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 4 3323.765 1633.834¤ 8067.418¤ 729.948 505.767 3754.100 18014.822 ¡ ¡ ¡ ¡ ¡ ¡ ¡ EF-1® 5 3317.822 1624.589 8044.044 731.392 505.767 3745.463 17969.078 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 2001 CATERINO ET AL.—PHYLOGENY OF SWALLOWTAILS 127

APPENDIX 3. Log likelihoods of rooted topologies for individual data partitions under two separate models. The best likelihoods under each partition and model are shown in boldface type. Asterisks indicate a likelihood signiŽcantly worse than the best for that partition (as determined by Kishino–Hasegawa test; ® 0:05): D COI/II EF-1®

Rooting Pos1 Pos2 Pos3 Pos1 Pos2 Pos3 Sum

HKY85 0 C 1 3719.239 1774.295 8884.742 815.482 534.946 4291.550 20020.254 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 2 3722.491 1774.746 8884.742 815.678 534.946 4291.720 20024.323 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 3 3722.516 1775.784 8899.366 815.678 534.946 4292.183 20040.471 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 4 3727.380 1782.155 8884.770 815.678 534.946 4298.015 20042.945 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 5 3727.877 1780.440 8899.420 815.678 534.946 4297.864 20056.224 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 6 3724.059 1774.746 8888.654 815.678 534.946 4291.687 20029.770

¡ ¡ ¡ ¡ ¡ ¡ ¡ Downloaded from 7 3722.561 1772.379 8888.654 808.371 534.946 4291.553 20018.460 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 8 3728.412 1785.538 8886.796 808.143 534.946 4293.163 20036.998 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 9 3731.059 1785.745 8886.796 808.177 534.946 4290.193 20036.916 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 10 3731.059 1785.384 8886.796 808.177 534.946 4293.681 20040.042 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 11 3732.306 1782.511 8886.796 812.805 534.946 4285.796 20035.160 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 12 3733.034 1782.511 8890.734 812.021 534.946 4285.896 20039.141

¡ ¡ ¡ ¡ ¡ ¡ ¡ http://sysbio.oxfordjournals.org/ 13 3731.450 1780.305 8890.734 805.289 534.946 4285.909 20028.634 ¡ ¡ ¡ ¡ ¡ ¡ ¡ GTR 0 C 1 3623.634 1755.207 8862.620 774.578 522.942 4224.652 19763.633 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 2 3626.978 1756.235 8862.620 775.259 522.942 4224.634 19768.668 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 3 3626.978 1756.635 8862.620 775.259 522.942 4225.449 19769.883 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 4 3630.080 1762.618 8862.655 775.259 522.942 4230.743 19784.296 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 5 3630.557 1760.820 8862.654 775.259 522.942 4230.690 19782.922 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 6 3628.259 1756.235 8866.699 775.259 522.942 4224.704 19774.098 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 7 3626.406 1753.676 8866.699 768.329 522.942 4224.574 19762.630 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 8 3628.800 1765.304 8864.384 767.315 522.942 4225.935 19774.679

¡ ¡ ¡ ¡ ¡ ¡ ¡ at Clemson University on June 18, 2014 9 3630.403 1766.265 8864.384 767.995 522.942 4224.176 19776.165 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 10 3630.701 1764.990 8864.384 767.995 522.942 4226.826 19777.839 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 11 3632.574 1763.144 8864.384 772.586 522.942 4219.662 19775.292 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 12 3632.835 1763.144 8895.457 ¤ 771.796 522.942 4219.803 19805.977 ¡ ¡ ¡ ¡ ¡ ¡ ¡ 13 3631.280 1760.744 8868.476 764.754 522.942 4219.705 19767.901 ¡ ¡ ¡ ¡ ¡ ¡ ¡